Process Knowledge: Can America Relearn How to Make Things?

23 Oct, 2026

TSMC is teaching Arizona to make chips America invented. For decades American companies taught Asia how to build what America designed. In Phoenix the lesson now runs the other way, and that makes it a key piece of US competitiveness in technology.

Dan Wang spent time in Phoenix this year talking to Taiwanese engineers, their families, and Arizona officials, and his essay in The New York Times on October 23 compresses what he found into one line: "TSMC is doing for America what Apple did for China." Wang's point is not about money or machines. Taiwan Semiconductor Manufacturing Company reports more than 3,500 employees in Arizona and an investment commitment that has grown from $12 billion in 2020 to $265 billion, but the asset Wang thinks matters most is what he has long called process knowledge, the accumulated practical skill that lets a factory turn a design into working product at volume. American companies spent decades exporting that skill to Asia. Taiwanese engineers on contracts of up to three years are now handing some of it back, one shift at a time.

I wrote about this place in June, asking why the world's most advanced chips were being made in a desert. That piece argued that Arizona's rise rested on human capital, on decades of Motorola and Intel engineering, on two large research universities, and on a state willing to bet its identity on knowledge rather than the copper, cattle, cotton, citrus, and climate that once defined it. Two days later I used the Golden Spider Silk Cape to argue that the technologies that change the world are not the most remarkable ones but the ones that become scalable. Wang's reporting lets me put those two essays together and ask the question they were circling. Attracting the factory was the easy part. Can the United States absorb the knowledge that makes the factory work?

TSMC Fab 21 under construction in Phoenix, Arizona, November 5, 2023. Photograph by Hunter Trick (TrickHunter). © Hunter Trick. Licensed under Creative Commons CC BY-SA 4.0. Source: Wikimedia Commons.

The split that made the foundry possible

Morris Chang spent 25 years at Texas Instruments, left in 1983, and in 1985 accepted Taiwan's invitation to run its Industrial Technology Research Institute. In 1987, with capital from the Taiwanese government and Philips, he founded TSMC as the world's first dedicated foundry, a factory that would manufacture chips designed by other companies and design none of its own. Intel, TI, and Motorola had all declined to invest. Chang's model institutionalized a separation between the people who design a chip and the people who make it, and that separation became the organizing principle of the modern industry. Apple, Nvidia, and Qualcomm design. TSMC builds. Chang himself, retired since 2018, has called America's attempt to reshore chipmaking "a very expensive exercise in futility," which is one more reason the Phoenix experiment deserves watching.

The separation worked so well that it hardened into a belief about value. Design was the high end, portable and well paid. Manufacturing was the commodity, something a country could shed without losing anything that mattered. Andy Grove, who had run Intel's factories, saw the flaw in 2010 when he answered Thomas Friedman's argument that Washington should back startups and let commodity manufacturing die. The phase after the garage, Grove wrote, is where companies work out the design details, learn to build affordably, and hire by the thousands. "Scaling is hard work but necessary to make innovation matter." Gary Pisano and Willy Shih made a similar case in the Harvard Business Review in 2009, describing an "industrial commons" of suppliers, process engineers, and skilled workers that erodes when production leaves, and taking most of the next generation of innovation with it.

Why a design is not finished until the fab says so

Wang has said he has spent much of the decade writing about the idea that design and manufacturing are one and the same thing. Here is why I think he is right, and why the design-versus-manufacturing frame breaks down at the level of the chip itself. A chip design is not a drawing that a factory then executes. Every foundry hands its customers a process design kit, a package of design rules, device models, and verification files that encodes what that particular process can and cannot reliably produce at that particular moment, and a design that violates those rules does not get built. Designers then revise as early silicon comes back with defects nobody predicted. Yield, the share of chips on a wafer that actually work, is manufacturing's verdict on the design, and a design that cannot be made at acceptable yield is not a finished design at all. The two activities are one activity observed from two ends. Chang's foundry model did not separate design from manufacturing. It separated the companies that do them, and left the knowledge that connects them sitting on the manufacturing side.

That knowledge is the kind Michael Polanyi had in mind in 1966 when he observed that we can know more than we can tell. Contamination control, tool calibration, the diagnosis of a yield excursion that appears on Tuesday and vanishes on Thursday, none of it survives transfer by document alone. Kenneth Arrow gave the economics of it a name in 1962, learning by doing, and the striking implication is that the learning decays when the doing stops. Wang has told the story of the United States forgetting how to make Fogbank, a classified material in thermonuclear warheads, because the engineers who understood it retired before anyone wrote the process down, and spending years recovering what it had once known. Nothing about a cleanroom is immune to the same forgetting.

What Arizona is actually doing about it

The encouraging news is that Arizona's institutions appear to understand that the transfer has to happen through people. Maricopa Community Colleges launched a ten-day Semiconductor Technician Quick Start program with Intel and TSMC in 2022 and certified 587 students in its first year, most of them students of color and more than half first-generation. Arizona State University announced a five-week Foundations for Equipment Technician program with TSMC in May, built in eight weeks, with a guaranteed interview at the end and more than 100 hires expected by the close of 2026. Those are modest numbers against a fab that will eventually need thousands, but they are the right shape. Technicians, not just engineers. Hands on tools, not just coursework. A pipeline that repeats.

The harder truth is that most of the Arizona ecosystem is still announced rather than built. TSMC's first fab has run 4-nanometer production since late 2024, but the second is only now receiving equipment, with 3-nanometer output targeted for the second half of 2027, and the third is under construction. Intel's Fab 52 in Chandler began high-volume 18A production in October 2025 and does not expect world-class yields until early 2027. Amkor's $2 billion packaging plant in Peoria, which would keep finished wafers from being shipped back to Asia for assembly, is a commitment with production expected in 2028, not a working factory. One operating fab does not make Taiwan less important in the near term, and nobody in Phoenix should pretend otherwise.

I have been following the Southwest as an innovation region for a long time, and I expect the problems to keep surfacing: water, housing, friction between a Taiwanese management culture built on long hours and an American workforce that expects something different, delays measured in quarters. Those problems would surface anywhere a region tried to build this fast. The engineers doing the teaching, by Wang's account, are in the country on temporary visas, which means the knowledge leaves with them unless it is absorbed first. The question that matters is whether the learning accumulates faster than it leaks.

Which brings me back to the premise. The conventional account of offshoring, which Wang describes as firms betting they could hold on to the more valuable work of research and development, says the United States kept the design and the intellectual property and the margins and let the low-value half go to Asia. On that account the loss was a trade, factory jobs given up and the brains retained. My argument is that the trade was never on offer, because design and fabrication were never two halves of anything. The knowledge that makes a design manufacturable lives on the fab floor, so sending the fab abroad did not preserve the half worth keeping. It kept the part of one activity that cannot be finished without the other part, and handed the finishing part to Taiwan. The United States did not lose chip manufacturing through cheap labor alone. It lost it because the people making the decision believed manufacturing was detachable, and the policy followed the belief. TSMC's engineers in Phoenix are not restoring a commodity. They are restoring the half of the design process that America stopped practicing, and the measure of success will not be how many fabs get built in the desert. It will be whether, a generation from now, the people teaching the next cohort in Arizona learned their craft there.

Further Reading

• Dan Wang, "The Factory Outside Phoenix That Could Save American Manufacturing," The New York Times, October 23, 2026. The reporting from Phoenix that prompted this essay.
• Why Are the World's Most Advanced Chips Being Made in a Desert? My June 21 essay on Arizona's human capital bet.
• Lessons from Silk: What Makes a Technology Scalable? My June 23 essay on the difference between invention and scalable production.
• Dan Wang, "How Technology Grows," 2018. Where Wang first laid out process knowledge as the hard-to-measure core of industrial capability.
• Andy Grove, "How America Can Create Jobs," Bloomberg Businessweek, July 1, 2010. Grove's argument that scaling, not invention, is where innovation becomes jobs.
• Gary Pisano and Willy Shih, "Restoring American Competitiveness," Harvard Business Review, July 2009. The "industrial commons" argument.
• Process design kit, Wikipedia. What a foundry gives its customers before any design can be manufactured.
• Michael Polanyi, The Tacit Dimension (1966), and Kenneth Arrow, "The Economic Implications of Learning by Doing," Review of Economic Studies 29, no. 3 (1962).
• TSMC Arizona. Company statement of investment, headcount, and fab milestones.
• ASU-TSMC program propels semiconductor careers, ASU News, October 7, 2026.
• Arizona's Leading Semiconductor Talent Pipeline, Arizona Commerce Authority, 2023. Source for the Quick Start first-year figures.

#AIData #Observations